Fast Spiking of a Mott VO2-Carbon Nanotube Composite Device

6 hours ago - This result provides an important component of energy-efficient neuromorphic computing systems, and a lithography-free technique for ...
0 downloads 0 Views 877KB Size
Subscriber access provided by Université de Strasbourg - Service Commun de la Documentation

Communication

Fast Spiking of a Mott VO2-Carbon Nanotube Composite Device Stephanie M. Bohaichuk, Suhas Kumar, Gregory Pitner, Connor J. McClellan, Jaewoo Jeong, Mahesh G. Samant, H.-S. Philip Wong, Stuart S. P. Parkin, R. Stanley Williams, and Eric Pop Nano Lett., Just Accepted Manuscript • DOI: 10.1021/acs.nanolett.9b01554 • Publication Date (Web): 21 Aug 2019 Downloaded from pubs.acs.org on August 21, 2019

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 14 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Nano Letters

Fast Spiking of a Mott VO2-Carbon Nanotube Composite Device Stephanie M. Bohaichuk1#, Suhas Kumar2*#, Greg Pitner1, Connor J. McClellan1, Jaewoo Jeong3, Mahesh G. Samant3, H-.S. Philip Wong1, Stuart S. P. Parkin3, R. Stanley Williams4, Eric Pop1,5 1Stanford

University, Electrical Engineering, Stanford, CA 94305, USA Packard Labs, 1501 Page Mill Rd, Palo Alto, CA 94304, USA 3IBM Almaden Research Center, 650 Harry Road, San Jose, CA 95120, USA 4Texas A&M University, Electrical & Computer Engineering, College Station, TX 77843, USA 5Stanford University, Material Science & Engineering, Stanford, CA 94305, USA 2Hewlett

*Address correspondence to S.K. ([email protected]). #Equal contribution

Abstract The recent surge of interest in brain-inspired computing and power-efficient electronics has dramatically bolstered development of computation and communication using neuron-like spiking signals. Devices that can produce rapid and energy-efficient spiking could significantly advance these applications. Here we demonstrate DC-current or voltage-driven periodic spiking with sub20 ns pulse widths from a single device composed of a thin VO2 film with a metallic carbon nanotube as a nanoscale heater, without using an external capacitor. Compared with VO2-only devices, adding the nanotube heater dramatically decreases the transient duration and pulse energy, and increases the spiking frequency, by up to three orders of magnitude. This is caused by heating and cooling of the VO2 across its insulator-metal transition being localized to a nanoscale conduction channel in an otherwise bulk medium. This result provides an important component of energy-efficient neuromorphic computing systems, and a lithography-free technique for energyscaling of electronic devices that operate via bulk mechanisms.

1 ACS Paragon Plus Environment

Nano Letters 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 2 of 14

The emergence of artificial intelligence and data-intensive tasks has necessitated a revamp of computing hardware beyond transistor-based Boolean logic and the von Neumann architecture.1,2 Within this revamping effort lies the broad domain of neuromorphic computing which aims to exploit biologically-inspired processes, namely computing, communicating, and operating a neural network using electrical spiking.3-8 In order to improve the energy-efficiency and speed of such systems it is desirable to control the pulse width and energy, and to produce the spiking using single scalable devices.4,9-11 For instance, adjusting the analog node weights of a neural network by small increments in order to enable high precision will require precise and tunable low energy pulses, especially in networks that use memristors such as phase change memory or oxide ionic resistive switches.12,13 Partly owing to the absence of compact circuits that can produce such tunable low-energy pulses, even the best memristor-based neural networks have had to implement elaborate transistorbased circuits at every node of very large networks, making the system’s efficiency far from ideal.14 Instead, compact spiking systems without transistors can be constructed by exploiting transient dynamics and/or electronic instabilities, for instance, the temporally abrupt resistance changes during a Mott insulator-metal transition (IMT) causing a capacitive discharge.15 VO2 and NbO2 are widely studied Mott insulators that undergo IMT above room temperature (~340 K in VO2, ~1100 K in NbO2), resulting in abrupt changes of their electrical resistance (usually by several orders of magnitude) often accompanied by measurable electrical instabilities such as negative differential resistance (NDR) as increasing current is applied.16,17 Relaxation oscillators that are constructed with NbO2 or VO2 typically use a DC-current or voltage driven PearsonAnson-like circuit that includes a capacitor in parallel with the NDR element, and a suitable series resistor.15,18-22 However, generation of fast DC-voltage-driven periodic spiking will require

2 ACS Paragon Plus Environment

Page 3 of 14 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Nano Letters

aggressive scaling to minimize both the electrical and thermal time constants of these oscillators. 23

Here we utilize a metallic single-wall carbon nanotube (CNT) of ~1 nm diameter as a nanoscale heater in contact with a thin film of VO2, in order to effectively scale a micrometersized device to sub-10 nm width,24 without requiring careful and time-consuming electron-beam lithography that would be impractical for constructing large arrays of neural networks. A single CNT-VO2 composite device with an integrated parallel capacitance from the contact electrodes forms a relaxation oscillator when driven by a DC source, producing periodic spiking. The effective scaling of the device width using a CNT yields dramatic improvements in the dynamic spiking behavior, including an increase in frequency and a reduction of pulse energy and transient time scales by nearly three orders of magnitude compared to VO2 control devices without a CNT. We further demonstrate that the frequency, duty cycle, and pulse width could be tuned within a single CNT-VO2 device by nearly one order of magnitude by altering the DC bias conditions. The device structure (Figures 1a-1c) contains a lateral active region defined by a ~5 nm thin strip of VO2 contacted by Pd electrodes, essentially forming a planar IMT device. To construct the CNT-VO2 composite device, aligned CNTs were grown on a quartz substrate then transferred onto the surface of VO2 before the device structure was defined (Figures 1d-1f) (see Supplementary Material for details of device fabrication and material properties). The quasi-static current-voltage behavior was measured using a current sweep (Figures 1g-1h) for both VO2 and CNT-VO2 devices. With an otherwise identical geometry (L = 5 I' W = 4 I'9 both exhibited qualitatively similar regions of NDR along with hysteresis. These behaviors are characteristic of a thermally-activated transport mechanism and a temperature-controlled Mott transition, detailed elsewhere.16,25 The quantitative differences between the two devices arise from a much narrower “thermal width” in

3 ACS Paragon Plus Environment

Nano Letters 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Paragon Plus Environment

Page 4 of 14

Page 5 of 14

introduction of a CNT. The simulations also show that the effective thermal capacitance reduced by 3 orders of magnitude. The thermal capacitance affects the dynamical behavior significantly more than the quasi-static characteristics, similar to how an electrical capacitance affects the dynamics more than the static behavior. The threshold quasi-static power to trigger NDR and switching was reduced by a factor of 2-5 in a CNT-VO2 compared to a VO2-only device24 while, as we will demonstrate, the dynamical behavior is affected by several orders of magnitude.

(a)

(b)

T (K) Pd electrode 328

Pd electrode

VO2

VO2

TiO2

296

TiO2

Pd electrode

CNT

Pd electrode

(c) T (K)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Nano Letters

(d) 3 nm

320 3.1 m

300 -2

296 K

-1

0 1 x ( m)

2 -2

-1

0 1 x ( m)

2

Figure 2: Simulated static temperature maps. Static surface temperature maps obtained by three-dimensional electrothermal finite element simulations of (a) a VO2-only device, and (b) a CNT-VO2 device at a DC voltage bias just below the switching threshold (both with L = 5 I' W = 4 I'95 Scale bar is 1 µm. (c)-(d) Temperature profiles along the dashed horizontal lines in (a) and (b), respectively. Blue dashed horizontal lines indicate room temperature. Red arrows indicate the width heated to 90% of the maximum temperature increase above the ambient, showing a reduction by three orders of magnitude in the CNT-VO2 device, compared to a VO2-only device. This measure is used to illustrate the significant reduction in the thermal mass upon introducing a CNT. To study the dynamical behavior of the two circuit elements, we applied a constant external current to each device within its NDR region. Because of the intrinsic capacitance of the devices and the large series resistance of the current source, a Pearson-Anson-like relaxation oscillator was 5 ACS Paragon Plus Environment

Nano Letters 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 6 of 14

formed. This produced a periodic train of repeating pulses or spikes, typical of Mott memristors, due to the periodic heating and cooling of VO2 across its IMT, which controls the periodic discharging and charging of the capacitor.27 Figures 3a-3b display the periodic spiking of a VO2only and a CNT-VO2 device that were nominally identical in geometry (L = 5 I' W = 4 I'9 except for the inclusion of the CNT in the latter. In the VO2-only device (Figures 3a, 3c) each spike consisted of an initial spike with a full width at half maximum (FWHM) of 28.5 ns (inset of Figure 3c), followed by a longer transient lasting 0.68 ms. Fitting an exponential decay to the falling edges yielded a time constant of 17 ns for the initial spike, and 0.24 ms for the longer transient (analyses detailed in Supplementary Material Figure S8). In contrast, the CNT-VO2 device (Figures 3b, 3d) showed a spike with a FWHM of 36.5 ns and single decay constant of 27 ns, with no apparent longer transient. Thus, although the CNT-VO2 device exhibited a spike comparable in duration to the initial large spike of the VO2-only device, the former exhibited nearly negligible transient dynamics following the spike. This resulted in a 1000-fold increase in frequency (~0.5 MHz in CNT-VO2 and 0.3 kHz in a VO2-only device), and a 100-fold reduction in total pulse energy (1.3 pJ in CNT-VO2 and 100 pJ in a VO2-only device). We used an indirect experimental technique (via polymer-coating of the devices, detailed in Supplementary Material Section 5) to qualitatively confirm that (a) the power dissipation and transient widths of the metallic phase of the VO2 were much smaller upon introduction of a CNT, (b) there were two time constants, a larger-amplitude faster spike and a smaller-amplitude slower transient, and (c) heating indeed occurred in the same location as the CNT.

6 ACS Paragon Plus Environment

Page 7 of 14

(a)

8

VO2-only device

CNT-VO2 device

(b)

vout (mV)

40

4 2

20

0 0

5

10 15 Time (ms)

0 0

20

(c) 8

vout (mV)

0 0

2

100 200 Time (ns)

0

(e)

10 15 Time ( s)

20

40

100

4

0.0

5

(d)

6

vout (mV)

Experimental data

vout (mV)

6

0.2

0.4 0.6 Time (ms)

0.8

20

0 0

1.0

100 Time (ns)

200

(f)

8

30

4 2 0 0

5

10

(h)

5

10 15 Time ( s)

20

30 100

6 0 0

4 2 0 0.0

20

0 0

20

10 8

vout (mV)

10 15 Time (ms)

0.2

vout (mV)

(g)

vout (mV)

vout (mV)

6

Simulated data

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Nano Letters

100 200 Time (ns)

0.4 0.6 Time (ms)

0.8

1.0

20 10 0 0

100 Time (ns)

200

Figure 3: Experimental and simulated periodic spiking. Measured periodic voltage spiking of (a) a VO2-only device driven by a 23 I DC current and (b) a CNT-VO2 device driven by a 60 I DC current, measured across the 50 O load of the oscilloscope. (c) Magnified plot of a single pulse in the VO2-only device, as indicated with a dashed rectangle in (a). The inset is a magnified plot of the transient spike at the rising edge of the pulse. (d) Magnified plot of a single pulse in the CNT-VO2 device, as marked with a dashed rectangle in (b), which is missing the long transient tail evident in (c). (e)-(h) are simulation results, in comparison with (a)-(d). All abscissa (time) values were arbitrarily offset for clarity of presentation. All insets have the same ordinate units as the corresponding panel. 7 ACS Paragon Plus Environment

Nano Letters 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 8 of 14

In order to gain insight into the effects of using a CNT to scale an electronic device down, we constructed a simplified compact device model that was incorporated into simulations of a relaxation oscillator circuit. The device model consisted of thermally-activated Schottky transport along with Newton’s law of cooling representing temperature dynamics, which together are known to produce instabilities such as NDR (detailed in the Supplementary Material Section 2).28,29 We used this model to reproduce the shape, frequency and the durations of the dynamics (Figures 3e3h) by altering only one parameter upon introduction of the CNT: the lumped thermal capacitance was reduced by about three orders of magnitude to account for the reduced volume of VO2 being heated across its IMT (consistent with the finite element simulations). The agreement of our model with experimental data is remarkable, given the simplicity of the model, and confirming our hypothesis that the CNT is indeed dramatically scaling down the effective thermal mass, the full effect of which can be inferred only by measuring the dynamics.

8 ACS Paragon Plus Environment

Page 9 of 14

(a)

2.5

100

100 k 200 k 300 k

10

20 30 Vext (V)

389 k

1.5 1.0

40

0.5 10

50

fmax

300 k 200 k

2.0 100 k

389 k

fmin 20

30 Vext (V)

40

50

(d) 2

10

f (MHz)

f (MHz)

200

0 0

(c)

(b) 3.0

20 k

fmax

E (pJ)

IRs ( A)

300

1

1 fmin 0

4 L ( m)

100

vout (mV)

(ns)

4 L ( m)

8

(f)

150

50 0 0

0 0

8

(e)

min

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Nano Letters

20

FWHM = 14 ns

0 4 L ( m)

8

0

25 Time (ns)

50

Figure 4: Periodic spiking behavior varying with device and circuit parameters. (a) Quasistatic current-voltage behavior of a single CNT-VO2 device with dimensions L = 1.8 I' and W = 6 I' measured using a voltage source and different series resistances, as marked in the colorcoded legend. (b) Oscillation frequencies resulting from biasing the device at specific voltages on the quasi-static behavior characteristics in (a). The highest frequencies occur at the highest voltages, but slightly shorter pulse widths occur at lower voltages and frequencies. (c)-(e) Dependence of maximum and minimum frequency, minimum energy delivered by a single pulse, and the minimum width (defined as the full width between 10% of the maximum) of a single pulse corresponding to varying lengths of different CNT-VO2 devices. (f) Plot of a single pulse within periodic spiking corresponding to the shortest device (L = 0.3 µm), at the minimum of the trends in (d) and (e).

9 ACS Paragon Plus Environment

Nano Letters 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 10 of 14

Within a single CNT-VO2 device, we studied the range of applied DC voltages and corresponding series resistances for which the device exhibited periodic spiking. As is expected from the theory of local activity and nonlinear dynamics,27 we observed that the range of voltages for which periodic spiking occurred increased with increasing series resistance (Figures 4a-4b), with the frequency range spanning nearly an order of magnitude. We further characterized the periodic spiking behavior across CNT-VO2 devices with varying lengths between the electrodes. The maximum and minimum frequencies, minimum pulse width, and the minimum energy delivered by a pulse, all spanned an order of magnitude as the device length was scaled by an order of magnitude (from 300 nm to 8 µm), displayed in Figures 4c-4e. In particular, the shortest device (L = 300 nm) exhibited the shortest pulse width, with a FWHM of ~14 ns, and a total energy delivered by the pulse (to a 50 O oscilloscope load) of ~93 fJ (Figure 4f). The shortest pulse width corresponded to the lowest of the measured frequencies within the device (~6 MHz). The shortest device also exhibited a maximum frequency of 11 MHz, the highest of all devices measured. The linear trend in the minimum pulse width (Figure 4e) had a non-zero y-intercept of ~15 ns, which likely represents the minimum possible pulse width within the present measurement setup and device structure (a combination of parasitic time constants and contact resistances). Additional analysis on the periodic spiking and the individual pulses (rise and fall times) are provided in the Supplementary Material Section 4. In conclusion, these results present a pathway to fast and low-energy neuron-like spiking with high demand in several types of neuromorphic computing applications. We have demonstrated that using a CNT nanoscale heater we can locally confine the effective thermal structure of an otherwise bulk behavior, thereby engineering the dynamical properties by several orders of magnitude without needing extensive and time-consuming nanoscale lithography. This provides a promising platform to scale and control the behavior of many electronic components which rely 10 ACS Paragon Plus Environment

Page 11 of 14 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Nano Letters

on temperature, given the increasing use of temperature-driven functions such as NDR, IMT, and chaos. Apart from the scaling of periodic spiking studied here, several other applications benefit from such thermal engineering, for instance within memory selectors which minimize the standby power consumption of data storage, while utilizing a thermal runaway to trigger threshold switching.30

Acknowledgements:

The authors gratefully acknowledge Kye Okabe, Gary A. Gibson, Saurabh Suryavanshi and Aditya Sood for helping with the experimental setup, simulation codes, calculations and/or commenting on the manuscript. This work was supported in part by the Stanford SystemX Alliance and by the National Science Foundation (NSF). Work was performed in part at the Stanford Nanofabrication Facility and the Stanford Nano Shared Facilities which receive funding from the National Science Foundation as part of the National Nanotechnology Coordinated Infrastructure Award ECCS1542152. S.B. acknowledges support from the Stanford Graduate Fellowship (SGF) program and the NSERC Postgraduate Scholarship program.

Supporting Information Available: The Supporting Information is available free of charge on the ACS Publications website at DOI: Operation of the Relaxation Oscillator, Compact Model used for Simulations, Device Fabrication and Experimental Setup, Additional Characterization of Oscillations, Using PMMA to Infer Temperature Dynamics, Three-Dimensional Finite Element Simulations, Comparison to smaller and crossbar devices, and endurance, Videos, LT Spice Model Files

11 ACS Paragon Plus Environment

Nano Letters 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 12 of 14

References 1. Williams, R. S. Comput. Sci. Eng. 2017, 19, 7-13. 2. Zidan, M. A.; Strachan, J. P.; Lu, W. D. Nat. Electron. 2018, 1, 22. 3. Pyle, R.; Rosenbaum, R. Phys. Rev. Lett. 2017, 118, 018103. 4. Wang, Z.; Rao, M.; Han, J.-W.; Zhang, J.; Lin, P.; Li, Y.; Li, C.; Song, W.; Asapu, S.; Midya, R. Nat. Commun. 2018, 9, 3208. 5. Merolla, P. A.; Arthur, J. V.; Alvarez-Icaza, R.; Cassidy, A. S.; Sawada, J.; Akopyan, F.; Jackson, B. L.; Imam, N.; Guo, C.; Nakamura, Y. Science 2014, 345, 668-673. 6. Yi, W.; Tsang, K. K.; K. Lam, S.; Bai, X.; Crowell, J. A.; Flores, E. A. Nat. Commun. 2018, 9, 4661. 7. Zhang, X.; Wang, W.; Liu, Q.; Zhao, X.; Wei, J.; Cao, R.; Yao, Z.; Zhu, X.; Zhang, F.; Lv, H. IEEE Electron Device Lett. 2018, 39, 308-311. 8. Lin, J.; Guha, S.; Ramanathan, S. Front. Neurosci. 2018, 12, 856. 9. Hsieh, C.-C.; Roy, A.; Chang, Y.-F.; Shahrjerdi, D.; Banerjee, S. K. Appl. Phys. Lett. 2016, 109, 223501. 10. Cao, Y.; Chen, Y.; Khosla, D. Int. J. Comp. Vis. 2015, 113, 54-66. 11. Del Valle, J.; Salev P.; Kalcheim, Y.; Schuller, I. K. arXiv 2019, 1903.01062 12. Sebastian, A.; Tuma, T.; Papandreou, N.; Le Gallo, M.; Kull, L.; Parnell, T.; Eleftheriou, E. Nat. Commun. 2017, 8, 1115. 13. Hu, M.; Graves, C. E.; Li, C.; Li, Y.; Ge, N.; Montgomery, E.; Davila, N.; Jiang, H.; Williams, R. S.; Yang, J. J. Adv. Mater. 2018, 30, 1705914. 14. Ambrogio, S.; Narayanan, P.; Tsai, H.; Shelby, R. M.; Boybat, I.; Nolfo, C.; Sidler, S.; Giordano, M.; Bodini, M.; Farinha, N. C. P. Nature 2018, 558, 60. 15. Pickett, M. D.; Medeiros-Ribeiro, G.; Williams, R. S. Nat. Mat. 2013, 12, 114-117. 16. Kumar, S.; Wang, Z.; Davila, N.; Kumari, N.; Norris, K.; Huang, X.; Strachan, J. P.; Vine, D.; Kilcoyne, A. L.; Nishi, Y.; Williams, R. S. Nat. Commun. 2017, 8, 658. 17. Kumar, S.; Pickett, M. D.; Strachan, J. P.; Gibson, G.; Nishi, Y.; Williams, R. S. Adv. Mater. 2013, 25, 6128-6132. 18. Pickett, M. D.; Williams, R. S. Nanotech. 2012, 23, 215202. 19. Li, S.; Liu, X.; Nandi, S. K.; Venkatachalam, D. K.; Elliman, R. G. Appl. Phys. Lett. 2015, 106, 212902. 20. Beaumont, A.; Leroy, J.; Orlianges, J. C.; Crunteanu, A. J. Appl. Phys. 2014, 115, 154502. 21. Shukla, N.; Parihar, A.; Freeman, E.; Paik, H.; Stone, G.; Narayanan, V.; Wen, H.; Cai, Z.; Gopalan, V.; Engel-Herbert, R.; Schlom, D. G.; Raychowdhury, A.; Datta, S. Sci. Rep. 2014, 4, 4964. 22. Kim, H.-T.; Kim, B.-J.; Choi, S.; Chae, B.-G.; Lee, Y. W.; Driscoll, T.; Qazilbash, M. M.; Basov, D. N. J. Appl. Phys. 2010, 107, 023702. 23. Kumar, S.; Strachan, J. P.; Williams, R. S. Nature 2017, 548, 318-321. 24. Bohaichuk, S. M.; Rojo, M. M.; Pitner, G.; McClellan, C. J.; Lian, F.; Li, J.; Jeong, J.; Samant, M. G.; Parkin, S. S. P.; Wong, H.-S. P.; Pop, E., ACS Nano 2019, DOI: 10.1021/acsnano.9b03397. 25. Funck, C.; Menzel, S.; Aslam, N.; Zhang, H.; Hardtdegen, A.; Waser, R.; Hoffmann Eifert, S. Adv. Electron. Mater. 2016, 2, 1600169. 26. Pop, E.; Mann, D. A.; Goodson, K. E.; Dai, H. J. Appl. Phys. 2007, 101, 093710. 27. Ascoli, A.; Slesazeck, S.; Mahne, H.; Tetzlaff, R.; Mikolajick, T. IEEE Trans. Circuits Syst. 2015, 62, 1165-1174. 12 ACS Paragon Plus Environment

Page 13 of 14 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Nano Letters

28. Gibson, G. A.; Musunuru, S.; Zhang, J.; Vandenberghe, K.; Lee, J.; Hsieh, C.-C.; Jackson, W.; Jeon, Y.; Henze, D.; Li, Z. Appl. Phys. Lett. 2016, 108, 023505. 29. Gibson, G. A. Adv. Funct. Mater. 2018, 28, 1704175. 30. Cha, E.; Park, J.; Woo, J.; Lee, D.; Prakash, A.; Hwang, H. Appl. Phys. Lett. 2016, 108, 153502.

13 ACS Paragon Plus Environment

Nano Letters 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Effective size-scaling of micrometer-sized VO2 Mott memristors using metallic carbon nanotube heaters speeds up the electrically-triggered Mott transition dynamics by nearly three orders of magnitude, thereby enabling aggressive energy-scaling of electronic components that utilize bulk material processes. 130x60mm (150 x 150 DPI)

ACS Paragon Plus Environment

Page 14 of 14